AIs Writing Academic Articles & Book Chapters

Many have started to notice the advances in the “Deep Researcher” tools from Google (Gemini), Perplexity, and OpenAI (ChatGPT-4o).

These tools have improved substantially over time, both in terms of their abilities to research academic content and in their abilities to “reason” through a signficant amount of it.

I’ve started writing a book chapter on AI pluralism and AI safety, so I thought I’d take each for a spin and see how they do.

Gemini Pro 2.5 Deep Research

I thought I’d start with this one, as there have been upgrades both to the quality of the Deep Research tool and the integration with the reasoning abilities of the 2.5 Pro model.

I do have a paid subscription to Gemini, but I did this using my own personal account/the free version.

This was the simple prompt I used.

Note: Yes, the prompt has typos, but AIs usually figure out what you want even with typos so I rarely fix them.

Gemini spent about 45 minutes “reading” through over 900 articles (I think the total count was 967).

It would show many of the sources it was “reading” as it went along. As I glanced over to see what it was doing, I’d say that all or nearly all of the sources were very credible.

It produced a 10 page “report” with 164 footnotes.

ChatGPT Deep Research

I entered the same prompt in ChatGPT-4o Deep Research that I entered into Gemini’s Deep Researcher.

When I put the prompt in, it did ask me some follow-up questions that I answered.

I chose 7,000 words beecause that was the lenght of the Gemini essay, even though I didn’t' specific a length in that case.

It produced a 24 page report with a 24 source bibliography.

Perplexity Deep Research

I entered the same prompt in Perplexity Deep Research that I entered into Gemini’s Deep Researcher.

Perplexity produced a 4 page overview with 39 sources.

Based on previous experience, I think Perplexity could do better than this. I may need to adjust the prompt. I didnt’ want to do that until I was done with this comparison.

How Will I Use This?

1. New ideas and sources I hadn’t thought about.

2. I will upload transcripts of discussions I’ve had with other authors of the book (it’s a collection) to include our collective ideas and thoughts for maintaining a consistent them and definitions of “pluralism” and “agents” throughout the book (as well as other concepts)

3. Thinking. With so much of the “grunt” research work out of the way, I’ll have a lot more time for thinking and collaboration.

4. Expression. With so much of the “grunt” research work out of the way, I’ll have a lot more time for thinking about how I want to express things. Of course, I’ll also ask AIs for help with that.

5. Don’t worry, yes, I’ll verify the sources.

6. Reading. I’ll have more time to read and think through the various articles that are in the produced bibiographies.

7. Some more research. When verifying sources, I’m sure I’ll find additional material that is relevant.

8. Sharing. I’ll have more time to collaborate with authors of other chapters.

9. Iterative refinement - using the initial research as a foundation, then having follow-up sessions with the AI tools to dive deeper into specific concepts or areas that need more exploration.

   Conclusion

   In the end, I think I’ll have a better product. My intelligence will be “amplified.”

A Couple Things I’m Curious About

1. Can I see the list of all 900+ sources that Gemini reviewed?

2. How did Gemini decide to exclude some of the sources?

Purpose

I originally put this together to simply show people the power of the Deep Research tools.

Beyond that, however, I think that over time I’ll share the process I used to produce the chapter, from this to the final product. Of course, you can see the final product in the book :).